Words and Rules; The Ingredients of Language; Stephen Pinker, Basic Books 1999


Reviewed by Graham Mulligan


The subject of this book is regular and irregular verbs. Everyone leans their native language in roughly the same way; lots of words and concepts assembled together following patterns and rules. People don’t just “blurt out words but rather combine them into phrases and sentences. So, what’s the issue with the verbs? Most of them (the regulars) behave similarly, for example add ‘ed’ to make their action in the past. But irregulars (there are about 150 – 180 of these all together) are really different, for example ‘go’ becomes ‘went’.


The irregular verbs have four particularly important words: ‘be’, ‘have’, ‘do’ and ‘go’ (corresponding to existence, possession, action and motion). These four are the most commonly used in most languages.


Pinker show us a neat diagrammatic way to understand the complexity of language (p.23):




Lexicon         Morphology         Syntax




Words, it turns out, should really be viewed as memorized chunks that the mind works with using rules for assembly. Things like idiomatic expressions (‘eat your heart out’ or ‘beat around the bush’), collocations (‘get away’), or clichés ( ‘time will tell’) are easily seen as chunks, but so is ‘walked’ (made up from ‘walk’ + ‘ed’).


Pinker dissects language into individual words, tracing how they have evolved through time. Verbs are particularly informative because they (usually) change from present tense to past tense in ways that can be linked to some past usage or form. The path involves Old English, Middle English, The Great Vowel Shift, Modern English and Americanization of English, as well as related languages like Old German and the Indo-European family of languages, all informing the words current existence. But we don’t need to know that history to use the words.


(Indo-European; Proto-Germanic; West Germanic (Angles and Saxons); Old English; Middle English; Modern English) Other families of languages that come from Indo-European: Germanic, Romance, Slavic, Celtic, Greek, Iranian and Sanskrit.


But is it so simple? Two competing theories see language formation very differently. Generative Phonology (Chomsky and Halle) describes language change as rule driven. Connectionism or Parallel Distributed Processing (Rumelhart and McClelland) sees associations between sounds of stems and past-tense forms generalized to new words similar to the old words. How do we think? This is an old debate between Rationalists (Descartes) and Empiricists (Hume). Is the mind packed with innate structure or is it a blank slate?


Pinker suggests the study of the past tense can illuminate this debate. Are the similarities between stems and their past tense forms (‘sing’ – ‘sang’; ‘drink’ – ‘drank’) following rules or are past tense forms memorized from some now defunct rule or fossil? Rumelhart and McClelland’s theory (1986) is described as an ‘input’ and ‘output’ model that works like a learning machine with ‘teacher’ input and repetition of the output until the correct form is learned. This is a brain theory using neural networks to explain how patterns of correct forms are then generalized to help learn new words as they are encountered. Going back to the diagram (p.23) this can now be represented as:





without the Lexical box or the Syntax box. This won’t do, according to Pinker, because the theory can describe ‘output’ or how to produce the correct sound (words) but doesn’t explain how the correct word can be recognized when first encountered. “Obviously people do both. Not only can we say ‘walked’, but when we hear ‘walked’ we know in means ‘walk’ in the past”. We learn rules and lexical entities so we can ‘send commands to the tongue’ and ‘interpret sounds coming in form the ear’. The pattern generator theory doesn’t account for verb stems.


Pinker concludes that a modified word and rule theory is closer to explaining this language puzzle, “regular forms are generated by rules, and irregular forms are retrieved from memory; the memory however, is not a list of slots but is partly associative, linking patterns with patterns as well as words with words”.


Diagram (p.118 & 152)


What is the main motor of productivity in language, rule processing or memory associations?


Consider this: a monster that eats a mouse every time he can is a ‘mice-eater’; a monster that eats a rat every time he can is a ‘rat-eater’ (not a ‘rats’ eater). ‘Mice’ is the plural form stored in memory. ‘Rats’ is the plural form from the rule: add ‘s’ to a stem to form a plural.

Leave a Reply

Your email address will not be published. Required fields are marked *